Technical Field
The present invention relates to an image recognition technology, and in particular, to a method for adsorbing a straight line/line segment in an image. The present invention further relates to an image recognition apparatus, and in particular, to an apparatus for adsorbing a straight line/line segment in an image. The present invention also relates to an image processing method, and in particular, to a method for constructing a polygon in an image. The present invention also relates to an image processing apparatus, and in particular, to an apparatus for constructing a polygon in an image.
Related Art
At present, an increasing quantity of functions of processing and retrieving information of a shot picture start emerging on a business device such as a scanner and a business phone. These functions may generally be categorized into two types:
1) Correction and Enhancement of a Whiteboard Image
A whiteboard image is an image obtained by shooting a conference whiteboard, a billboard and other text-type rectangular targets. Information in which a user is interested in a whiteboard image is generally in a rectangular area. In a picture obtained through shooting, first there is an unwanted edge area outside a target rectangular area, and next because of a problem of a shooting angle, a case of an irregular quadrilateral (as shown in FIG. 1) or a missing corner (as shown in FIG. 2) may occur in the target rectangular area. An image is processed to remove parts outside a target area and transform an irregular quadrilateral into a regular rectangle (as shown in FIG. 3), or to fill a missing corner part (as shown in FIG. 4); a color image is converted into a black-and-white image, and an image enhancement operation is performed to make text clearer.
2) Recognition of Text in a Text Image
A text image is an image obtained by shooting a standard text, for example, an A4-format paper text, a page in various books, and a movie poster. In text recognition, data of a text image is processed by using a text recognition engine to recognize text in the image. Text content in a text image is generally displayed in a rectangular area. In a picture obtained through shooting, there is an unwanted edge area outside a target rectangular area, and next, because of a problem of a shooting angle, a case of an irregular quadrilateral (as shown in FIG. 1) or a missing corner (as shown in FIG. 2) may occur in the target rectangular area. To enable a text recognition engine to recognize a text image more accurately and rapidly, the image needs to be processed to remove parts outside a target area and transform an irregular quadrilateral into a regular rectangle (as shown in FIG. 3), or to fill a missing corner part (as shown in FIG. 4); and a color text image is converted into a black-and-white image, and an image enhancement operation is performed to make a contrast between text and the background in the image clearer.
In both types of operations above, a quadrilateral frame detection operation needs to be performed. The quadrilateral frame detection operation refers to that detection is performed on a shot image to retrieve a target quadrilateral frame, so that in a subsequent operation, an area outside the quadrilateral frame is removed and an operation of rectangular frame correction or corner filling is performed on the quadrilateral frame. However, in an actual operation, a range of a quadrilateral frame obtained through automatic detection may be not an expected result. As shown in FIG. 5, the part of the dotted-line range is supposed to be a target frame, but the part of the solid-line range is actually detected; or as shown in FIG. 6, in a source image, there are multiple frames that can be used for capturing. The part of the dotted-line range on the left side is supposed to be a target frame, but instead the part of the solid-line range on the right side is actually detected. Therefore, during initial detection of an image, one interaction interface needs to be provided to a user to adjust a target frame.
In the disclosed text of Chinese Invention Patent Publication No. CN101807293A, a method for constructing a quadrilateral frame by using vertex adsorption is recorded. However, in an actual operation, because a shooting background is relatively messy, it may be difficult to recognize a position of a vertex, or a vertex of a quadrilateral to be recognized is a rounded corner. These all result in that a vertex adsorption function becomes unusable, and a user cannot construct a quadrilateral frame by using a vertex.